# Song Lab — What makes it a 10?

Evidence-backed research, not a guaranteed perfect-song formula. Five owner-liked references, 18 dimensions and 12 falsifiable hypotheses.

## Reference songs

- [[references/krov]] — krov
- [[references/media-922fbf8f0e9c]] — media-922fbf8f0e9c
- [[references/sisterbrazil]] — sisterbrazil
- [[references/svekr]] — svekr
- [[references/targ]] — targ

## Concepts

- [[concepts/story]] — Story momentum
- [[concepts/density]] — Syllable load
- [[concepts/release]] — Phrase endings
- [[concepts/pulse]] — Pulse & accents
- [[concepts/motif]] — Returning musical figure
- [[concepts/contour]] — Melodic contour
- [[concepts/harmony]] — Harmonic color
- [[concepts/carrier]] — Accompaniment roles
- [[concepts/development]] — Arrangement development
- [[concepts/foreground]] — Voice proximity
- [[concepts/voice]] — Vocal identity
- [[concepts/emotion]] — Emotional posture
- [[concepts/source]] — Audio conditioning
- [[concepts/style-weight]] — Style adherence
- [[concepts/weirdness]] — Creative deviation
- [[concepts/persona]] — Persona route
- [[concepts/entrance]] — Lyrical entrance
- [[concepts/ending]] — Ending & breathing space

## Hypotheses

- [[experiments/H1]] — H1 · Phrase motion beats raw word speed
- [[experiments/H2]] — H2 · A recognizable figure carries the song
- [[experiments/H3]] — H3 · Source influence has a sweet spot
- [[experiments/H4]] — H4 · Persona stabilizes identity, not taste
- [[experiments/H5]] — H5 · Constant density is an overconstraint
- [[experiments/H6]] — H6 · A 1.25-second entrance gate may be too blunt
- [[experiments/H7]] — H7 · Fewer instructions can produce more music
- [[experiments/H8]] — H8 · Meaningful contrast beats more hits or loudness
- [[experiments/H9]] — H9 · Reviewer confidence is not a taste label
- [[experiments/H10]] — H10 · Voice gender, exact key and a lyric chorus are lane choices
- [[experiments/H11]] — H11 · Local syllable hotspots predict weak phrases
- [[experiments/H12]] — H12 · One BPM cannot define the groove

## Executed local experiments

- [[experiments/E01]] — E01 · Constant word speed is not the shared secret.
- [[experiments/E02]] — E02 · Returning music does not require returning lyrics.
- [[experiments/E03]] — E03 · A single BPM gate loses the groove question.
- [[experiments/E04]] — E04 · A small entrance relaxation changes nothing here.
- [[experiments/E05]] — E05 · Two repeated stalls deserve an operator diagnosis.
- [[experiments/E06]] — E06 · Most historical revisions cannot isolate a knob.
- [[experiments/E07]] — E07 · A formal pass can miss a local lyric defect.
- [[experiments/E08]] — E08 · Find heavy phrases before asking for more speed.

## Method

[[SCHEMA]] · [[concepts/validation]] · [[concepts/contradictions]] · [[log]]

## Open questions

Which generated songs does the owner rate 9–10? Which single intervention wins repeatedly? What source/style-weight region works? Does Persona improve preference or just consistency?